texture_cache: Use small vector for surface vectors
This avoids most heap allocations when collecting surfaces into a vector.
This commit is contained in:
parent
508242c267
commit
d2b2557542
@ -14,6 +14,7 @@
|
|||||||
#include <unordered_map>
|
#include <unordered_map>
|
||||||
#include <vector>
|
#include <vector>
|
||||||
|
|
||||||
|
#include <boost/container/small_vector.hpp>
|
||||||
#include <boost/icl/interval_map.hpp>
|
#include <boost/icl/interval_map.hpp>
|
||||||
#include <boost/range/iterator_range.hpp>
|
#include <boost/range/iterator_range.hpp>
|
||||||
|
|
||||||
@ -53,6 +54,7 @@ using RenderTargetConfig = Tegra::Engines::Maxwell3D::Regs::RenderTargetConfig;
|
|||||||
|
|
||||||
template <typename TSurface, typename TView>
|
template <typename TSurface, typename TView>
|
||||||
class TextureCache {
|
class TextureCache {
|
||||||
|
using VectorSurface = boost::container::small_vector<TSurface, 1>;
|
||||||
|
|
||||||
public:
|
public:
|
||||||
void InvalidateRegion(VAddr addr, std::size_t size) {
|
void InvalidateRegion(VAddr addr, std::size_t size) {
|
||||||
@ -498,7 +500,7 @@ private:
|
|||||||
* @param untopological Indicates to the recycler that the texture has no way
|
* @param untopological Indicates to the recycler that the texture has no way
|
||||||
* to match the overlaps due to topological reasons.
|
* to match the overlaps due to topological reasons.
|
||||||
**/
|
**/
|
||||||
RecycleStrategy PickStrategy(std::vector<TSurface>& overlaps, const SurfaceParams& params,
|
RecycleStrategy PickStrategy(VectorSurface& overlaps, const SurfaceParams& params,
|
||||||
const GPUVAddr gpu_addr, const MatchTopologyResult untopological) {
|
const GPUVAddr gpu_addr, const MatchTopologyResult untopological) {
|
||||||
if (Settings::IsGPULevelExtreme()) {
|
if (Settings::IsGPULevelExtreme()) {
|
||||||
return RecycleStrategy::Flush;
|
return RecycleStrategy::Flush;
|
||||||
@ -538,9 +540,8 @@ private:
|
|||||||
* @param untopological Indicates to the recycler that the texture has no way to match the
|
* @param untopological Indicates to the recycler that the texture has no way to match the
|
||||||
* overlaps due to topological reasons.
|
* overlaps due to topological reasons.
|
||||||
**/
|
**/
|
||||||
std::pair<TSurface, TView> RecycleSurface(std::vector<TSurface>& overlaps,
|
std::pair<TSurface, TView> RecycleSurface(VectorSurface& overlaps, const SurfaceParams& params,
|
||||||
const SurfaceParams& params, const GPUVAddr gpu_addr,
|
const GPUVAddr gpu_addr, const bool preserve_contents,
|
||||||
const bool preserve_contents,
|
|
||||||
const MatchTopologyResult untopological) {
|
const MatchTopologyResult untopological) {
|
||||||
const bool do_load = preserve_contents && Settings::IsGPULevelExtreme();
|
const bool do_load = preserve_contents && Settings::IsGPULevelExtreme();
|
||||||
for (auto& surface : overlaps) {
|
for (auto& surface : overlaps) {
|
||||||
@ -650,7 +651,7 @@ private:
|
|||||||
* @param params The parameters on the new surface.
|
* @param params The parameters on the new surface.
|
||||||
* @param gpu_addr The starting address of the new surface.
|
* @param gpu_addr The starting address of the new surface.
|
||||||
**/
|
**/
|
||||||
std::optional<std::pair<TSurface, TView>> TryReconstructSurface(std::vector<TSurface>& overlaps,
|
std::optional<std::pair<TSurface, TView>> TryReconstructSurface(VectorSurface& overlaps,
|
||||||
const SurfaceParams& params,
|
const SurfaceParams& params,
|
||||||
const GPUVAddr gpu_addr) {
|
const GPUVAddr gpu_addr) {
|
||||||
if (params.target == SurfaceTarget::Texture3D) {
|
if (params.target == SurfaceTarget::Texture3D) {
|
||||||
@ -708,7 +709,7 @@ private:
|
|||||||
* @param preserve_contents Indicates that the new surface should be loaded from memory or
|
* @param preserve_contents Indicates that the new surface should be loaded from memory or
|
||||||
* left blank.
|
* left blank.
|
||||||
*/
|
*/
|
||||||
std::optional<std::pair<TSurface, TView>> Manage3DSurfaces(std::vector<TSurface>& overlaps,
|
std::optional<std::pair<TSurface, TView>> Manage3DSurfaces(VectorSurface& overlaps,
|
||||||
const SurfaceParams& params,
|
const SurfaceParams& params,
|
||||||
const GPUVAddr gpu_addr,
|
const GPUVAddr gpu_addr,
|
||||||
const VAddr cpu_addr,
|
const VAddr cpu_addr,
|
||||||
@ -810,7 +811,7 @@ private:
|
|||||||
TSurface& current_surface = iter->second;
|
TSurface& current_surface = iter->second;
|
||||||
const auto topological_result = current_surface->MatchesTopology(params);
|
const auto topological_result = current_surface->MatchesTopology(params);
|
||||||
if (topological_result != MatchTopologyResult::FullMatch) {
|
if (topological_result != MatchTopologyResult::FullMatch) {
|
||||||
std::vector<TSurface> overlaps{current_surface};
|
VectorSurface overlaps{current_surface};
|
||||||
return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
|
return RecycleSurface(overlaps, params, gpu_addr, preserve_contents,
|
||||||
topological_result);
|
topological_result);
|
||||||
}
|
}
|
||||||
@ -1124,14 +1125,14 @@ private:
|
|||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|
||||||
std::vector<TSurface> GetSurfacesInRegion(const VAddr cpu_addr, const std::size_t size) {
|
VectorSurface GetSurfacesInRegion(const VAddr cpu_addr, const std::size_t size) {
|
||||||
if (size == 0) {
|
if (size == 0) {
|
||||||
return {};
|
return {};
|
||||||
}
|
}
|
||||||
const VAddr cpu_addr_end = cpu_addr + size;
|
const VAddr cpu_addr_end = cpu_addr + size;
|
||||||
VAddr start = cpu_addr >> registry_page_bits;
|
VAddr start = cpu_addr >> registry_page_bits;
|
||||||
const VAddr end = (cpu_addr_end - 1) >> registry_page_bits;
|
const VAddr end = (cpu_addr_end - 1) >> registry_page_bits;
|
||||||
std::vector<TSurface> surfaces;
|
VectorSurface surfaces;
|
||||||
while (start <= end) {
|
while (start <= end) {
|
||||||
std::vector<TSurface>& list = registry[start];
|
std::vector<TSurface>& list = registry[start];
|
||||||
for (auto& surface : list) {
|
for (auto& surface : list) {
|
||||||
|
Loading…
Reference in New Issue
Block a user